Question regarding R520 and R580

Radeon600

Newcomer
I have a question in my mind. I understand Dave's Article on X1800 Series, in which he stated that RV530 might be similar to R580 Architectural wise, which means most probably 'Rodin' or R580 will have 24 Shader Units with 48 Shader ALUs and 192 ALUs total. But, I don't understand one thing, doesn't R520 has 2 Shader ALUs (ALU1 & ALU2) per Fragment Processor also? so, doesn't this mean it has 32 Shader ALUs in total per Pixel Pipeline? or there's is single Shader ALU in single Shader Unit?

Thanks everyone.
 
afaik one of these ALU's is "mini" ie can do only scalar math(1 component per op), while the "full" ALU is vector (4 component per op)
 
http://www.beyond3d.com/reviews/ati/r520/index.php?p=03

It seems one ALU is "reduced", but I wouldn't call it mini.

Wonder what this bodes for Xenos? It seems to have 48 full strength ALU's.

Perhaps this is also why on a clock per pipe basis, Nvidia is trumping ATI. They seem to have two full ALU's per pipe.

For example a 6800 GS beats a X800 GTO despite similar pipe counts and clockspeeds.
 
That means there's only one 'Powerfull' Shader ALU per Pixel Pipeline present in R520? So, in 7800GTX there are two Shader ALUs per Pixel Pipeline supported by 'Mini ALUs', doesn't it mean it has 4 Shader ALUs per Pixel Pipeline with 96 Shader ALUs in Total?
 
Radeon600 said:
I have a question in my mind.
You've got some strange numbers there, so think of it like this. R520 has 16 fragment pipelines, each pipeline has one pixel shader unit, each pixel shader unit has two ALUs (one full, one mini). So you end up with 32 ALUs. RV530 has four fragment pipelines, each pipeline has three pixel shader units, each pixel shader units has two ALUs (again, one full, one mini). Here, 24 ALUs. The rumour for R580 is that it's effectively four times RV530; 16 fragment pipelines, 3 pixel shader units per pipeline, 2 ALUs per shader unit, so 96 ALUs.

Edit: My bad on the G70 bit. What's the go with it's mini ALUs?
 
Last edited by a moderator:
So R580 will be roughly double Xenos.

This is why MS should have waited and used R580.

And come out at the same time as PS3.
 
Radeon600 said:
I am little confused with the following Pictorial representation of 7800GTX Shader Unit
Ah, that's confused me a little too. An important thing to consider however is that like many other things, ALUs are not comparable between different architectures.
 
chavvdarrr said:
afaik one of these ALU's is "mini" ie can do only scalar math(1 component per op), while the "full" ALU is vector (4 component per op)

No, thats Xenos. Each of the fragment shaders on R5xx consists of two vector ALU's (that can each co-issue up to Vec3 and Scalar), however one of them has the full capabilities while the other is just a vector ADD unit.
 
Bill said:
So R580 will be roughly double Xenos.

This is why MS should have waited and used R580.

And come out at the same time as PS3.
Eh? Xenos has got a whole pile of features that won't be seen until R600. Why would XB360 benefit from a less capable GPU?

Jawed
 
Alright, I have confused myself with those numbers once more. In my first post I mentioned that R580 will have 192 ALUs and 7800GTX already has 192 ALUs of total, in all present Shader Units.

If there are two Shader ALUs in single Shader Unit, and if R580 also has 2 'FULL' Shader Unit so it makes it a total of 96 ALUs but not 192 ALUs?


4 Shader ALUs per Shader Unit = 24 x 4 = 96 ALUs?


That's:

24 Vector1 ALUs
24 Vector2 ALUs
24 Scalar1 ALUs
24 Scalar2 ALUs


Thanks
 
:?:

Shouldn't it be:

2 ALUs/shader unit
x 3 shader units/"pipe"
x 16 "pipes" = 96 PS ALUs

48 Vec3 + Scalar ADD (/+imput modifier?)
48 Vec3 + Scalar ADD/MULL/MADD

and then vertex shaders...
 
Bill said:
ALU wise.

But they'll be higher clockspeed, improved memory controller, etc. so you're right, more like 2.5 :)

Don't be silly a R580 would could most likely be clocked around 400MHz at best to fit the power envelope desired in the XB360, it would cost way more to produce than the r500/c1 and would also increase cost by needing aditional traces and memory chips for a 256bit interface - this would not only increase the production cost at launch but the whole lifespan of the console.
 
Just wondering about one more thing, if the 7800GTX has two 'FULL' Shader ALUs supported with Mini ALU per Shader Unit, then why Nvidia says its 24 Unit chip?, logically doesn't it means it has 48 Shader Units?

R580 likely to have two Quads, with 3 Independent Shader Units (All DOT?) per Shader Unit.
 
Radeon600 said:
Just wondering about one more thing, if the 7800GTX has two 'FULL' Shader ALUs supported with Mini ALU per Shader Unit, then why Nvidia says its 24 Unit chip?, logically doesn't it means it has 48 Shader Units?

R580 likely to have two Quads, with 3 Independent Shader Units (All DOT?) per Shader Unit.

Well my understanding is that there is a difference between the number of fragment shaders and the number of ALU's per fragment shader.
 
Well, what that difference would be?.. I cannot understand because there's MADD operation in both Shader ALU, and hence they act like 2 Independent Shader Units, especially from ATI point of view?
 
Last edited by a moderator:
Back
Top